What deep learning adds

What you’ll learn

This is the opening lesson of Track 12 (Introduction to Deep Learning) and it sits directly on top of the previous track. If you know what a neural network is (layers of neurons, weights tuned by gradient descent and backpropagation), this lesson takes that engine and asks the bigger questions: what does “deep” add, why did it suddenly start working, and what is the whole field actually made of? The source curriculum is MIT 6.S191 by Alexander and Ava Amini, freely available at introtodeeplearning.com.

The lesson opens with a puzzle (the core ideas are decades old, so why did deep learning only take over around 2012?), answers it with the depth-data-compute story and the 2012 AlexNet moment, uses the XOR example to show why depth matters at all, previews the four problem shapes the track will tour, and closes with an honest account of what deep learning is not.

Where this fits

This is lesson 1 of 10, and the entry point of the track. There is no previous lesson here; the prerequisite is the Neural Network Intuition track, which built the engine this whole survey assumes. The next lesson, Why sequences need memory, begins the tour proper with the first problem shape (ordered data), and the rest of the track works through vision, generation, and decisions before closing on the honest limits and a synthesis.

Before you start

Prerequisites: ideally the previous track (Neural Network Intuition), or equivalent comfort with what a neural network is and how it learns. If “layers, weights, gradient descent, backpropagation” all feel familiar, you are ready. If they do not, that track is the place to firm them up first; this lesson deliberately does not re-teach them.

About the math

Track 12 is a survey, and this opening lesson needs no math at all. It is conceptual: the “why now” story, the XOR intuition (shown with a picture and a hands-on browser demo, no equations), and a map of the field. Later lessons in the track stay at the intuition level too, with the occasional small worked example you can follow without a calculator.

By the end, you’ll be able to

Explain what the word “deep” adds to a neural network (many layers plus the training tricks that make depth work)
Explain why deep learning took off around 2012 rather than the 1980s, in terms of depth, data, and compute arriving together
Use the XOR example to explain why depth lets a network build complex patterns from simple ones
Name the four problem shapes the track surveys (sequences, images, generation, decisions) and recognize that one engine underlies all four
State honestly what deep learning is not, and why it is both powerful and bounded

Time and difficulty

Read time: about 9 minutes
Practice time: about 10 minutes (a hands-on TensorFlow Playground demo that makes depth matter with your own eyes, plus flashcards)
Difficulty: standard (a conceptual orientation lesson; no math)